Analysis and Application of English Intelligent Translation Error Detection Model based on HITS Algorithm

In response to the semantic discontinuity and other problems existing in the current English intelligent translation, this paper proposes a graph neural network error detection model based on the HITS algorithm. First, a dependency tree is generated based on Stanford CoreNLP, and the initial authority value of the node is set as the TF-IDF weight; then a dynamic propagation rule is designed to calculate the syntactic relevance between nodes through a three-layer graph convolutional network; then the BERT fine-tuning mechanism is introduced for semantic calibration, and the node representation is updated synchronously in each HITS iteration; finally, a decision module containing authority-hub dual thresholds is constructed. When the node authority value exceeds 0.78 and the hub value is lower than 0.35, it is judged as an error point. Tests on the WMT20 dataset show that this method has achieved an accuracy of 93.2% and an F1 value of 89.6% in English-Chinese translation tasks. This model effectively captures the topological propagation characteristics of translation errors and provides a new technical path for intelligent translation quality control.